Skip to content
Open access

Gated Cross-Modal Fusion and Feature Reconstruction for Visible-Infrared Remote Sensing Object Detection

2026 · IEEE Transactions on Geoscience and Remote Sensing · Vol 64, pp. 5409818-5409818 · 1 citation · 81 references

Abstract

The visible-infrared object detection in uncrewed aerial vehicle (UAV) remote sensing remains challenging due to small object feature degradation, complex-background interference, modality heterogeneity, and the high computational cost of existing fusion methods. To address these issues, this study proposes a multimodal gated interaction and reconstruction network (MGIR-Net). First, a feature reconstruction module (FRM) recalibrates intramodal representations by enhancing local saliency, suppressing directional background interference, and compensating for global response bias. Second, a cross-modal gated interaction module (CGIM) performs Gaussian error linear unit (GELU)-based gated modulation on fused visible-infrared features to enhance reliable complementary cues while suppressing unreliable responses. Third, a hierarchical contextual guidance module (HCGM) enlarges the effective receptive field through decoupled spatial sampling, providing spatial priors for small object localization with limited overhead. The effectiveness and generalization ability of MGIR-Net are demonstrated by extensive experiments on two aerial remote sensing datasets, DroneVehicle and VEDAI, and four complementary visible-infrared benchmarks, LLVIP, M3 FD, MFAD, and KAIST. The proposed method achieves mAP50 scores of 85.5%, 78.6%, 96.5%, 90.5%, 79.0%, and 75.2% on the six datasets, respectively. With 6.9-M parameters and 19.3 GFLOPs, MGIR-Net provides a favorable balance between detection accuracy and computational efficiency. The source code for this work is publicly available at https://github.com/Leon1998314/MGIR-Net.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.